Update ocr_engine.py

ocr_engine.py

@@ -32,25 +32,25 @@ def estimate_brightness(img):
     return np.mean(gray)
 
 def preprocess_image(img):
-    """Preprocess image for OCR with aggressive contrast and noise reduction."""
+    """Preprocess image with aggressive contrast and noise handling."""
     gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
     brightness = estimate_brightness(img)
-    # Aggressive CLAHE
-    clahe_clip = 6.0 if brightness < 80 else 4.0
-    clahe = cv2.createCLAHE(clipLimit=clahe_clip, tileGridSize=(8, 8))
+    # Maximum CLAHE for extreme contrast
+    clahe_clip = 10.0 if brightness < 80 else 6.0
+    clahe = cv2.createCLAHE(clipLimit=clahe_clip, tileGridSize=(6, 6))
     enhanced = clahe.apply(gray)
     save_debug_image(enhanced, "01_preprocess_clahe")
-    # Minimal blur to preserve edges
-    blurred = cv2.GaussianBlur(enhanced, (3, 3), 0)
+    # Edge-preserving blur
+    blurred = cv2.bilateralFilter(enhanced, 5, 75, 75)
     save_debug_image(blurred, "02_preprocess_blur")
-    # Multi-scale thresholding
-    block_size = max(9, min(25, int(img.shape[0] / 20) * 2 + 1))
+    # Adaptive thresholding with small blocks
+    block_size = max(5, min(15, int(img.shape[0] / 30) * 2 + 1))
     thresh = cv2.adaptiveThreshold(blurred, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C, 
-                                   cv2.THRESH_BINARY_INV, block_size, 7)
-    # Morphological operations
+                                   cv2.THRESH_BINARY_INV, block_size, 3)
+    # Morphological operations for digit segmentation
     kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (3, 3))
-    thresh = cv2.morphologyEx(thresh, cv2.MORPH_OPEN, kernel, iterations=2)
-    thresh = cv2.morphologyEx(thresh, cv2.MORPH_CLOSE, kernel, iterations=3)
+    thresh = cv2.morphologyEx(thresh, cv2.MORPH_OPEN, kernel, iterations=1)
+    thresh = cv2.morphologyEx(thresh, cv2.MORPH_CLOSE, kernel, iterations=5)
     save_debug_image(thresh, "03_preprocess_morph")
     return thresh, enhanced
 
@@ -58,12 +58,12 @@ def correct_rotation(img):
     """Correct image rotation using edge detection."""
     try:
         gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
-        edges = cv2.Canny(gray, 30, 100, apertureSize=3)
-        lines = cv2.HoughLinesP(edges, 1, np.pi / 180, threshold=40, minLineLength=20, maxLineGap=10)
+        edges = cv2.Canny(gray, 20, 80, apertureSize=3)
+        lines = cv2.HoughLinesP(edges, 1, np.pi / 180, threshold=30, minLineLength=15, maxLineGap=5)
         if lines is not None:
             angles = [np.arctan2(line[0][3] - line[0][1], line[0][2] - line[0][0]) * 180 / np.pi for line in lines]
             angle = np.median(angles)
-            if abs(angle) > 0.5:
+            if abs(angle) > 0.3:
                 h, w = img.shape[:2]
                 center = (w // 2, h // 2)
                 M = cv2.getRotationMatrix2D(center, angle, 1.0)
@@ -76,20 +76,20 @@ def correct_rotation(img):
         return img
 
 def detect_roi(img):
-    """Detect region of interest with aggressive contour filtering."""
+    """Detect region of interest with flexible contour filtering."""
     try:
         save_debug_image(img, "04_original")
         thresh, enhanced = preprocess_image(img)
         brightness_map = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
-        block_sizes = [max(9, min(25, int(img.shape[0] / s) * 2 + 1)) for s in [10, 15, 20]]
+        block_sizes = [max(5, min(15, int(img.shape[0] / s) * 2 + 1)) for s in [6, 10, 15]]
         valid_contours = []
         img_area = img.shape[0] * img.shape[1]
         
         for block_size in block_sizes:
             temp_thresh = cv2.adaptiveThreshold(enhanced, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C, 
-                                                cv2.THRESH_BINARY_INV, block_size, 7)
+                                               cv2.THRESH_BINARY_INV, block_size, 3)
             kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (5, 5))
-            temp_thresh = cv2.morphologyEx(temp_thresh, cv2.MORPH_CLOSE, kernel, iterations=3)
+            temp_thresh = cv2.morphologyEx(temp_thresh, cv2.MORPH_CLOSE, kernel, iterations=5)
             save_debug_image(temp_thresh, f"05_roi_threshold_block{block_size}")
             contours, _ = cv2.findContours(temp_thresh, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
             
@@ -98,15 +98,15 @@ def detect_roi(img):
                 x, y, w, h = cv2.boundingRect(c)
                 roi_brightness = np.mean(brightness_map[y:y+h, x:x+w])
                 aspect_ratio = w / h
-                if (500 < area < (img_area * 0.5) and 
-                    0.5 <= aspect_ratio <= 6.0 and w > 80 and h > 40 and roi_brightness > 60):
+                if (200 < area < (img_area * 0.7) and 
+                    0.2 <= aspect_ratio <= 10.0 and w > 50 and h > 20 and roi_brightness > 40):
                     valid_contours.append((c, area * roi_brightness))
                     logging.debug(f"Contour (block {block_size}): Area={area}, Aspect={aspect_ratio:.2f}, Brightness={roi_brightness:.2f}")
         
         if valid_contours:
             contour, _ = max(valid_contours, key=lambda x: x[1])
             x, y, w, h = cv2.boundingRect(contour)
-            padding = max(25, min(70, int(min(w, h) * 0.5)))
+            padding = max(15, min(40, int(min(w, h) * 0.3)))
             x, y = max(0, x - padding), max(0, y - padding)
             w, h = min(w + 2 * padding, img.shape[1] - x), min(h + 2 * padding, img.shape[0] - y)
             roi_img = img[y:y+h, x:x+w]
@@ -122,66 +122,98 @@ def detect_roi(img):
         save_debug_image(img, "06_roi_error_fallback")
         return img, None
 
-def detect_digit_contour(digit_img, brightness):
-    """Simplified contour-based digit recognition."""
+def detect_digit_template(digit_img, brightness):
+    """Digit recognition using template matching with predefined patterns."""
     try:
         h, w = digit_img.shape
-        if h < 20 or w < 10:
-            logging.debug("Digit image too small for contour detection.")
+        if h < 10 or w < 5:
+            logging.debug("Digit image too small for template matching.")
             return None
 
-        # Normalize image
-        pixel_count = np.sum(digit_img == 255)
-        total_pixels = digit_img.size
-        density = pixel_count / total_pixels
-        if density < 0.1 or density > 0.8:
-            return None
-
-        # Contour analysis
-        contours, _ = cv2.findContours(digit_img, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
-        if not contours:
-            return None
+        # Predefined digit templates (simplified binary patterns)
+        digit_templates = {
+            '0': np.array([[1, 1, 1, 1, 1],
+                           [1, 0, 0, 0, 1],
+                           [1, 0, 0, 0, 1],
+                           [1, 0, 0, 0, 1],
+                           [1, 1, 1, 1, 1]]),
+            '1': np.array([[0, 0, 1, 0, 0],
+                           [0, 0, 1, 0, 0],
+                           [0, 0, 1, 0, 0],
+                           [0, 0, 1, 0, 0],
+                           [0, 0, 1, 0, 0]]),
+            '2': np.array([[1, 1, 1, 1, 1],
+                           [0, 0, 0, 0, 1],
+                           [1, 1, 1, 1, 1],
+                           [1, 0, 0, 0, 0],
+                           [1, 1, 1, 1, 1]]),
+            '3': np.array([[1, 1, 1, 1, 1],
+                           [0, 0, 0, 0, 1],
+                           [1, 1, 1, 1, 1],
+                           [0, 0, 0, 0, 1],
+                           [1, 1, 1, 1, 1]]),
+            '4': np.array([[1, 0, 0, 0, 1],
+                           [1, 0, 0, 0, 1],
+                           [1, 1, 1, 1, 1],
+                           [0, 0, 0, 0, 1],
+                           [0, 0, 0, 0, 1]]),
+            '5': np.array([[1, 1, 1, 1, 1],
+                           [1, 0, 0, 0, 0],
+                           [1, 1, 1, 1, 1],
+                           [0, 0, 0, 0, 1],
+                           [1, 1, 1, 1, 1]]),
+            '6': np.array([[1, 1, 1, 1, 1],
+                           [1, 0, 0, 0, 0],
+                           [1, 1, 1, 1, 1],
+                           [1, 0, 0, 0, 1],
+                           [1, 1, 1, 1, 1]]),
+            '7': np.array([[1, 1, 1, 1, 1],
+                           [0, 0, 0, 0, 1],
+                           [0, 0, 0, 0, 1],
+                           [0, 0, 0, 0, 1],
+                           [0, 0, 0, 0, 1]]),
+            '8': np.array([[1, 1, 1, 1, 1],
+                           [1, 0, 0, 0, 1],
+                           [1, 1, 1, 1, 1],
+                           [1, 0, 0, 0, 1],
+                           [1, 1, 1, 1, 1]]),
+            '9': np.array([[1, 1, 1, 1, 1],
+                           [1, 0, 0, 0, 1],
+                           [1, 1, 1, 1, 1],
+                           [0, 0, 0, 0, 1],
+                           [1, 1, 1, 1, 1]]),
+            '.': np.array([[0, 0, 0],
+                           [0, 1, 0],
+                           [0, 0, 0]])
+        }
 
-        contour = max(contours, key=cv2.contourArea)
-        x, y, cw, ch = cv2.boundingRect(contour)
-        if cw < 5 or ch < 10:
-            return None
-
-        aspect = cw / ch
-        area_ratio = cv2.contourArea(contour) / (cw * ch)
-
-        # Simplified digit patterns
-        if aspect > 0.2 and aspect < 0.4 and area_ratio > 0.5:
-            return '1'
-        elif aspect > 0.5 and area_ratio > 0.6:
-            if density > 0.5:
-                return '8'
-            elif density > 0.3:
-                return '0'
-        elif aspect > 0.4 and area_ratio > 0.5:
-            if density > 0.4:
-                return '3'
-            elif density > 0.3:
-                return '2'
-        elif aspect > 0.3 and area_ratio > 0.4:
-            return '5' if density > 0.3 else '7'
-        elif aspect > 0.2 and area_ratio > 0.3:
-            return '4' if density > 0.2 else '9'
-        return None
+        # Resize digit_img to match template size (5x5 for digits, 3x3 for decimal)
+        digit_img_resized = cv2.resize(digit_img, (5, 5), interpolation=cv2.INTER_NEAREST)
+        best_match, best_score = None, -1
+        for digit, template in digit_templates.items():
+            if digit == '.':
+                digit_img_resized = cv2.resize(digit_img, (3, 3), interpolation=cv2.INTER_NEAREST)
+            result = cv2.matchTemplate(digit_img_resized, template, cv2.TM_CCOEFF_NORMED)
+            _, max_val, _, _ = cv2.minMaxLoc(result)
+            if max_val > 0.7 and max_val > best_score:
+                best_score = max_val
+                best_match = digit
+        logging.debug(f"Template match: {best_match}, Score: {best_score:.2f}")
+        return best_match if best_score > 0.7 else None
     except Exception as e:
-        logging.error(f"Contour digit detection failed: {str(e)}")
+        logging.error(f"Template digit detection failed: {str(e)}")
         return None
 
 def perform_ocr(img, roi_bbox):
-    """Perform OCR with Tesseract and contour-based fallback."""
+    """Perform OCR with Tesseract and template-based fallback."""
     try:
         thresh, enhanced = preprocess_image(img)
         brightness = estimate_brightness(img)
         pil_img = Image.fromarray(enhanced)
         save_debug_image(pil_img, "07_ocr_input")
         
-        # Tesseract with aggressive numeric config
-        custom_config = r'--oem 3 --psm 7 -c tessedit_char_whitelist=0123456789.'
+        # Tesseract with flexible numeric config
+        custom_config = r'--oem 3 --psm 6 -c tessedit_char_whitelist=0123456789.'
         text = pytesseract.image_to_string(pil_img, config=custom_config)
         logging.info(f"Tesseract raw output: {text}")
         
@@ -192,17 +224,17 @@ def perform_ocr(img, roi_bbox):
         text = text.strip('.')
         if text and re.fullmatch(r"^\d*\.?\d*$", text):
             text = text.lstrip('0') or '0'
-            confidence = 98.0 if len(text.replace('.', '')) >= 3 else 95.0
+            confidence = 97.0 if len(text.replace('.', '')) >= 3 else 94.0
             logging.info(f"Validated Tesseract text: {text}, Confidence: {confidence:.2f}%")
             return text, confidence
 
-        # Fallback to contour-based detection
-        logging.info("Tesseract failed, using contour-based detection.")
+        # Fallback to template-based detection
+        logging.info("Tesseract failed, using template-based detection.")
         contours, _ = cv2.findContours(thresh, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
         digits_info = []
         for c in contours:
             x, y, w, h = cv2.boundingRect(c)
-            if w > 15 and h > 20 and 0.2 <= w/h <= 1.2:
+            if w > 8 and h > 10 and 0.1 <= w/h <= 2.0:
                 digits_info.append((x, x+w, y, y+h))
         
         if digits_info:
@@ -216,10 +248,10 @@ def perform_ocr(img, roi_bbox):
                     continue
                 digit_crop = thresh[y_min:y_max, x_min:x_max]
                 save_debug_image(digit_crop, f"08_digit_crop_{idx}")
-                digit = detect_digit_contour(digit_crop, brightness)
+                digit = detect_digit_template(digit_crop, brightness)
                 if digit:
                     recognized_text += digit
-                elif x_min - prev_x_max < 15 and prev_x_max != -float('inf'):
+                elif x_min - prev_x_max < 8 and prev_x_max != -float('inf'):
                     recognized_text += '.'
                 prev_x_max = x_max
             
@@ -229,8 +261,8 @@ def perform_ocr(img, roi_bbox):
             text = text.strip('.')
             if text and re.fullmatch(r"^\d*\.?\d*$", text):
                 text = text.lstrip('0') or '0'
-                confidence = 92.0 if len(text.replace('.', '')) >= 3 else 90.0
-                logging.info(f"Validated contour text: {text}, Confidence: {confidence:.2f}%")
+                confidence = 92.0 if len(text.replace('.', '')) >= 3 else 89.0
+                logging.info(f"Validated template text: {text}, Confidence: {confidence:.2f}%")
                 return text, confidence
         
         logging.info("No valid digits detected.")
@@ -240,18 +272,18 @@ def perform_ocr(img, roi_bbox):
         return None, 0.0
 
 def extract_weight_from_image(pil_img):
-    """Extract weight from a digital scale image."""
+    """Extract weight from any digital scale image."""
     try:
         img = np.array(pil_img)
         img = cv2.cvtColor(img, cv2.COLOR_RGB2BGR)
         save_debug_image(img, "00_input_image")
         img = correct_rotation(img)
         brightness = estimate_brightness(img)
-        conf_threshold = 0.9 if brightness > 100 else 0.7
+        conf_threshold = 0.8 if brightness > 100 else 0.6
 
         roi_img, roi_bbox = detect_roi(img)
         if roi_bbox:
-            conf_threshold *= 1.15 if (roi_bbox[2] * roi_bbox[3]) > (img.shape[0] * img.shape[1] * 0.3) else 1.0
+            conf_threshold *= 1.05 if (roi_bbox[2] * roi_bbox[3]) > (img.shape[0] * img.shape[1] * 0.2) else 1.0
 
         result, confidence = perform_ocr(roi_img, roi_bbox)
         if result and confidence >= conf_threshold * 100:
@@ -266,7 +298,7 @@ def extract_weight_from_image(pil_img):
 
         logging.info("Primary OCR failed, using full image fallback.")
         result, confidence = perform_ocr(img, None)
-        if result and confidence >= conf_threshold * 0.95 * 100:
+        if result and confidence >= conf_threshold * 0.85 * 100:
             try:
                 weight = float(result)
                 if 0.01 <= weight <= 1000: